#407 – Guillaume Verdon: Beff Jezos, E/acc Movement, Physics, Computation & AGI
Lex Fridman Podcast
Tendency Towards Decentralization of AI - Hierarchical Systems Work Better
The current trend in AI leans towards centralization due to compute density and data centralization. However, there is a prediction of running out of available internet data. Efforts are being made to increase compute density for distributed information acquisition and hypothesis testing. The idea of a fundamentally centralized cybernetic control with one massive intelligence is considered suboptimal. Hierarchical cybernetic control, similar to nature and corporations, is presented as a more optimal approach. It involves individual contributors with self-interests, management with a wider scope, and larger timescales, creating a fault-tolerant system. This hierarchical and fault-tolerant structure is comparable to quantum error correction with local decoders performing the best. The prediction is for a correction towards decentralization of AI, breaking it into smaller subsystems that can communicate with each other and form a meta-system.
Remember Everything You Learn from Podcasts
